topic Re: Qlik Cloud AutoML Timeseries in Qlik Predict

Qlik Cloud AutoML Timeseries

JacobForssell — Mon, 26 Sep 2022 14:58:37 GMT

Hi!

I've been trying to create a timeseries forecast modell using Qlik AutoML but i cant get it to work properly.

In my data i have two columns which are sales and date. From what i have understood, AutoML should automatically give me the option to choose a Time series model given that i have a date column. Instead i only get regression models a suggestions

source:

(https://support.bigsquid.com/hc/en-us/articles/360023985793-SONAR-Time-Series-in-Qlik-AutoML ) (https://support.bigsquid.com/hc/en-us/articles/222760748-How-are-algorithms-chosen ).

Does someone know what could be wrong/ if its even possible to create a timeseries model atm?

Re: Qlik Cloud AutoML Timeseries

KellyHobson — Mon, 26 Sep 2022 15:25:36 GMT

Hey @JacobForssell

Timeseries was deprecated from Qlik AutoML (on Qlik Cloud) as there is the same capability with built natively in the Qlik Sense line chart.

Here is more information about the Qlik Sense line chart.

Let me know if you have any additional questions.

Best,

Kelly

Re: Qlik Cloud AutoML Timeseries

fabianweareactive — Tue, 27 Sep 2022 08:48:14 GMT

Hi @KellyHobson

I was following this topic, since I had the same problem. Due to the deprecation, it seems that a few functionalities have been removed.

1. Is it possible to compare the forecasted values with the actual values?

2. Is it possible to export the forecasted values to a file or shown them in a table?

Thanks!

Re: Qlik Cloud AutoML Timeseries

paulcalvet — Tue, 27 Sep 2022 15:24:21 GMT

Hello, I'm agree, forecast in line chart is limited.

We have moore possibilities when we precalculate forecast and the data are really available in the application.

Regards,

Re: Qlik Cloud AutoML Timeseries

Kyle_Jourdan — Tue, 27 Sep 2022 17:35:50 GMT

There are plan to potentially bring a better timeseries capability to AutoML in the future.

In the meantime, if the timeseries functionality in the line chart is not sufficient for your use case, it is still possible to manually engineer timeseries-type features and build a regression experiment in AutoML today.

As a starting point, any labels extracted from a date that may provide seasonality guidance (month, week of year, etc.) as well as rolling windows of the target (year over year, average past 6 months, etc.) would be a few suggestions to start that process. It will certainly take a little more data prep, but will provide for that more robust forecasting experience.

Re: Qlik Cloud AutoML Timeseries

fabianweareactive — Wed, 28 Sep 2022 06:05:42 GMT

Hi, thank you for your answer. Can you provide me a link to the webpage that shows an example of this precalculation?

Thanks in advance.

Re: Qlik Cloud AutoML Timeseries

Kyle_Jourdan — Wed, 28 Sep 2022 16:57:07 GMT

A quick Google search led to this article on the topic:

https://medium.com/data-science-at-microsoft/introduction-to-feature-engineering-for-time-series-forecasting-620aa55fcab0

I'd encourage you to utilize resources such as Medium, Toward Data Science, and Google to find some of the many examples of doing this type of feature engineering.

Re: Qlik Cloud AutoML Timeseries

MikkoL — Fri, 29 Sep 2023 05:20:44 GMT

1 and 2 fair problems since we have to use other tools for forecasting and to be fair when using other tools they start to take place over the non fuctional tool (qlik).

Re: Qlik Cloud AutoML Timeseries

LwamB — Mon, 22 Apr 2024 11:33:06 GMT

I am working with sensor data and I built a regression model in AutoML, but now I am having a hard time having it forecast values for the future (for example, I want to see what the values of my readings will be in the upcoming month). How can I go about this issue?

Thanks in advance!

Re: Qlik Cloud AutoML Timeseries

Kyle_Jourdan — Mon, 22 Apr 2024 16:52:03 GMT

Hello @LwamB –

In order to forecast into the future, you will need to create “synthetic” rows of data for those future dates.

For example, if your data contains fields such as “ProductCategory” and “LocationName”, you need to create a cartesian join between the future date values (create one row for each month in the future you want to forecast) and all the features in your model.

If you were creating a forecast for 3 months in the future, you would have a row for every unique combination of “ProductCategory” and “LocationName” multiplied by 3 (for each future month date you create).